A question about the actions of Large Players

[Pascal]

Pascal,

This is a think-piece question, not urgent or even all that important for any trades or on-going business, but a speculation based on this thread, that I've been pondering while watching the tape, and starting to gather some data for:

The effective volume is volume which moves price within a minute. The move may be as small as $0.01, or even smaller. I recall that the size of the intra-minute price change is not addressed in the calculations - and that's the way I'm doing it. A move of $.25 has the same weight as a move of $0.0075. In both cases, the volume moved the price, so it counts as effective volume.

Now, I've studied the book's false-positive cases, and I've had a couple of those myself. MFB was one I posted at the old forum - it took over a quarter to escape from that one. But more interesting (to me, anyway) are the price pops which come out of a clear blue sky, with no warning, no LEV, no SEV, no nothing visible. Which led me to wonder - the book is over 3 years old. It's taught in some business schools, and I recall you even directly address fund managers about how to conceal their movements in one of the later chapters. What if they listened? After all, the structural problems of large fund accumulation have not changed.

So I've been trying to postulate how to create a notional algorithm specifically designed to conceal effective volume as ineffective.

1. Starting with the prior day's tape, and the current day's evolution, I would calculate the equi-power volume level. That variable would be re-set as needed, perhaps even multiple times per day. That level becomes my per-trade volume ceiling. Never trade more than that in a single minute.

2. For each trade, I would set the lot size to be ([variable 1.] - [a randomized percentage of the current offer])

3. Within the algo there would have to be a countdown clock which resets to 59 seconds each minute, starting on the first second of the open. Trades would start at some second above 0, based on how many shares to target that day, divided by the 390 trading minutes, then played out against the actual liquidity, but always leveled so as to have the greatest chance of spanning a price level accumulation across the minute boundary. For example:

If this were the snapshot at +3 seconds...

Bid Shares| Shares Ask
20.00 1000 900 20.01
20.00 200 200 20.01
20.00 100 2000 20.01
19.99 500 1600 20.02

... the algo would buy 800 @ +2 seconds, then 100 @ + 1 seconds, then 2100 at +59 seconds of the next minute. Presuming, of course, the offers stayed stable. This would be happening at hyper speed, so the algo would be dynamically sensitive to such changes, polling the book several hundred times per second. In each trade burst, a portion of the offered lot would be left for someone else to clean up, so as to extend the odds of leaving a price level "tail" across the minute boundary.

As long as price levels are traded across the minute boundary, and with the equi-power boundary governing lot size, this algo's activity becomes invisible to the EV formula, and its volume becomes ineffective volume.

As a pseudo-Excel formula, isn't it true that Ineffective volume could be identified by reversing the volume detection formula from:

=IF(price now= price previous, 0,(((ABS(price previous-price now)+ABS(price now-price previous)) / (high now-low now+ABS(close now-close previous))) * volume now) * IF(close now-close previous>0,1,-1) )

into:

=IF(price now= price previous, (((ABS(price previous-price now)+ABS(price now-price previous)) / (high now-low now+ABS(close now-close previous))) * volume now) * IF(close now-close previous>0,1,-1),0 )
?

Or, more simply, reverse the true and false actions; make the "true part" of the formula the volume calculation, and make the "false part" of the formula zero - the result, I think, will return the ineffective volume?

The goal of such a survey would be to sample a set of high-volume stocks with flat or declining LEV, and look for a steadily rising large ineffective volume flow compared to a prior look-at period, and see if that is similarly associated to later price moves, as with effective volume. Or, stated as a motive - the purpose is to try to detect whether anyone has specifically attempted to cloak themselves from the EV detector.

It is no great effort - I do EVs by the dozen most days anyway. But has anyone thought about something like this, for the vast majority of stocks which lack strong LEV most of time whose prices pop with no warning?

This could be a great research work Ellis. Do you intend to perform it?
I for sure do not have the tools right now to do that.

Pascal

[Pascal]

Pascal,

Thank you for the detailed answer to Nickola's question. I have similar questions but not having a strong trading background don't have as good an idea of where to start.

I would certainly agree that "smart" players would carefully hide their transactions. Perhaps my primary question about EV methodology is why the large players (who we assume have all means at their disposal to hide their intentions) would cause a large change in price from minute to minute (compared to the high-low difference during that minute), which is what LEV/SEV separation is based on. Wouldn't they want to structure their transactions to produce a small price change at a high volume?

I think I am asking why you chose the Williams method as the basis for EV, as opposed to, for example, an OBV of some form, or even something like looking for the smallest price change at high volume as potential evidence for "big smart" transactions. Is there prior work that shows some basis for this model? I would certainly be interested in learning about the issues. Did you try other methods of detecting EV before writing your book?

Anyway, it may very well be that the details of the low level calculations are not as important as the averaging effect of 1000+ stocks.

The selection of the Larry Williams' adapted method is what looked best as I did not have access to tick data at that time. What I however did was to use as effective volume the total volume traded during one minute, as far as the minute was showing a price inflection. This method (OBV) was also good at shwoing accumulation/distribution. Larry Williams seemed to be more precise though. Today, I'd use tick data for sure instead of LW.



Pascal

[grems8544]

The actual method could be improved by shortening the analysis time interval and by using tick data instead of the Larry Williams formula to count shares within one time interval.

I think this is the key in the next evolution of EV.

Ellis' comments notwithstanding, I've noticed many trades which are large-block trades do not budge the TEV lines, or do so disproportionately. We certainly could argue (successfully) that over that 1 minute period that the equilibrium of the buyers and sellers did not change, but when I step back and look at a large block transacting, *something* is going on with the stock, yet TEV will not reflect the massive buying or selling. As an example of this I provide GRT's action on Friday:



Here, we see that over 3M shares transacted in a single minute, yet LEV/TEV did not budge. Due to scaling you cannot see the volume action after this -- there were numerous smaller blocks transacting (2K-10K size), which moved LEV/TEV around. Certainly, the 3M transaction did not upset the equilibrium, but when I step back and look at the stock as a whole, the 3M represents a $30M transaction by *somebody*, and I would want to know which way this moving (a buy or a sell).

Another example is when a large purchase barely moves TEV, but smaller volume moves TEV a considerable amount. As an example of this is COMV on 5/23:



Here, we see that early in the trading day, a considerable amount of a volume transaction occurs (bottom trace, red), and LEV (blue) and TEV (yellow) (middle traces) are barely moving. Later in the day, we have significant movement in LEV and TEV, yet the relative volume levels are much lower.

I think that in both these examples that volume vs. TICK information would be more indicative of the bigger picture for the stock, not just the equilibrium change.

My three cents (inflation) ....

pgd

[Pascal]

Paul,


There is a large block filtration algo inside of the method. It is not explained in the VIT book, but since a large block that does not impact the price is probably an "arranged" trade (for example between the MM and his customer), you want to filter this out, otherwise this sort of transaction "pollutes" the EV pattern. The filtration algo is to filter out the minutes whose traded volume is higher than 5% of the daily volume, but which does not impact the price by more than 1%.

[grems8544]

Thanks Pascal. I need to ponder this .... :o)

[TraderD]

I find it interesting (but not really surprising) that the EV method discussion gets directed towards tick-based volume analysis. Linnsoft's Investor/RT tool includes "Volume Breakdown" and "Cumulative Delta" capabilities for several years (and instructional videos on their site: http://www.linnsoft.com/vb) as well as courses offered by Fulcrum Trader (http://www.fulcrumtrader.com, previously active on EliteTrader.com under AMT4SWA nickname). I had some hands-on experience with system testing using VB principles in 2009 but the statistical edge using a single instrument turned out to be too slim to be of practical use for retail trading.

The VB analysis is reportedly most relevant to instruments that trade 24x7 and capture all traded volume (e.g. ES futures contract on Globex). Maybe this is where discrepancies in EV measurements produce less-than-ideal examples on a stock-by-stock basis. My own hunch (tho not proven) is that EV analysis shines due to the law of large numbers when statistics is aggregated across a large number of stocks over a period of days/weeks. In other words, if you look under a microscope, you're bound to find plenty of errors, but taken together, you get a much better signal/noise indicator.

Availability of tick-data history (unless you're a GS) is rather limited and would require considerable storage and data management efforts to carry out, compared with processing 1-min data. Saying it, if the EV method accuracy warrants it and can be translated into a non-trivial boost to EV robot performance, such effort would be quite worthwhile IMHO.

-TraderD

[Andrei]

FWIW, here is a company that provides historical tick data for equities, futures, options and so on...

http://www.tickdata.com/

[Ellis Wyatt]

This could be a great research work Ellis. Do you intend to perform it?
I for sure do not have the tools right now to do that.

Pascal

I don't have tools of the quality the robot, or the ability to backtest multiple years, or the discipline to do this at the level which would satisfy most members to risk their own money on the results; having said that, I have the tools to produce my own active boundary and EV flow charts in Excel, at the one-minute resolution, and I'm pretty sure the results match those given by the methods described in the book. I also concur, and I think many members have had the thought, that the tick is the final destination, though retaining that much data for enough symbols to be meaningful is beyond my current capacity.

Initially, what I am doing is just taking a given stock's EV spreadsheet, copying it, inverting the logic of the EV detection formula by copying it down the OHLCV columns, and observing the results. So far I don't have enough data to risk any observations, but when I do, I'll post them up. The reason I'm willing to try this at all is that everyone's computer trading systems, as a feature, are based on clocks network-synchronized to the second world-wide, which means that everyone's one-minute charts are very likely looking at the exact same single minutes. It's that technological presumption that makes the idea of a cross-minute algo hiding in plain sight seem so devilishly clever - no one in their right mind would intentionally skew their own trading clock, and only institutions have the horsepower and staff to chart, store, and analyze tick data on samples and timelines large enough to trust. And as many already know, day traders and longer-hold investors generally prefer longer timeframes than one minute, rather than shorter.

I've been out of the cubicle world for quite a while, but if I were going to try to conceal large accumulation or distribution from EV detectors, that's how I'd do it, because its so relatively cheap and easy and based on a presumption that no one thinks to mess with.

[Ellis Wyatt]

Pascal,

The current equities range obviously has us independent amateurs confused. I'm wondering if you have ever given any thought to intervals in which the large players act similarly confused? The below image reminded me of the book's IMAX example:



FITB is a regional US bank, not special in any particular way. They paid back the TARP, but nothing material has changed since the last earnings, which missed @ $0.10 v. expected $0.26 and lower revenues, on April 21.


The IMAX example was a case of false-positive, where value investors rush in, turn out to have mis-timed, and are then followed by bottom feeders (who may also turn out to be early). Have you, or anyone, seen this double-clutch pattern before in a case where there is no news of significance?

(I'm not buying, nor do I hold this name. But I've seen a couple of these double head-bumps lately, and they seem new. Unless it's just that I'm relatively new with EV and haven't seen enough permutations yet.)

[Pascal]

Pascal,

The current equities range obviously has us independent amateurs confused. I'm wondering if you have ever given any thought to intervals in which the large players act similarly confused? The below image reminded me of the book's IMAX example:

Attachment 8592

FITB is a regional US bank, not special in any particular way. They paid back the TARP, but nothing material has changed since the last earnings, which missed @ $0.10 v. expected $0.26 and lower revenues, on April 21.


The IMAX example was a case of false-positive, where value investors rush in, turn out to have mis-timed, and are then followed by bottom feeders (who may also turn out to be early). Have you, or anyone, seen this double-clutch pattern before in a case where there is no news of significance?

(I'm not buying, nor do I hold this name. But I've seen a couple of these double head-bumps lately, and they seem new. Unless it's just that I'm relatively new with EV and haven't seen enough permutations yet.)

This is a good question for which there is no good answer.
Statistically, if you buy in the direction of an increasing 3D EV and sell X-days later, the returns are lower than a Buy/Hold strategy. This means that EV has no predictive value when taken independently. (That observation was published in the VIT book). What is important is to use EV at the right timing: where the market/sector or stock hesitates, such as in a trading range or at a support/resistance level, or at an Active Boundaries limit.

FITB is really heavily traded. So we might have some funds buying and other funds selling, which means "no clear view" from large players.